JP2005071115A - Registration/retrieval method for object in p2p environment and program therefor - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a registration/retrieval method in a P2P environment that is compatible with efficient object retrieval with the use of various feature vectors, load equalization, node addition and deletion. <P>SOLUTION: A desired object is retrieved by using a common retrieval method utilizing signatures from the objects registered according to a common registration method utilizing each signature in nodes of maximum 2<SP>n</SP>having a node-ID specifying itself and connected to a network, respectively. In both registration and retrieval methods, locators are generated. Index entries are generated on the basis of the locators, and the index entries are registered in the nodes equal to or more than one that are specified by the locators. Index entries are collected on the basis of inquiry locators, and all index entries satisfying conditions of inquiry frame signatures are transmitted to nodes that perform retrieval work. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to an object registration / retrieval method using a signature in a P2P environment, and a program used to implement this method.

  In recent years, Peer-to-Peer (P2P) technology has attracted attention due to the high performance of computers and the development of network infrastructure. In P2P, each computer (computer) becomes a peer node (node) to construct a large-scale distributed network. Each node plays both roles without distinction between servers and clients. The P2P network is classified into a hybrid P2P type in which client / server systems are integrated and a pure P2P type that is a completely distributed environment. In the hybrid P2P type, a specific server exists in order to provide a certain type of service. In the object search in the hybrid P2P type, it is possible to share information distributed to the nodes by providing an index function by the server. However, when a large number of nodes request the service of the server, the server becomes a bottleneck. Further, when the server is stopped, the entire service is stopped. On the other hand, in the pure P2P type, each node operates autonomously. For this reason, it is rich in extensibility such that a node can be added without being restricted by the processing capability of the server, and processing without a bottleneck can be realized.

  However, since the pure P2P type cannot have a global index or the like in the server, information sharing is generally not easy. As a typical pure P2P type system, there is a file sharing system Gnutella [Non-patent Document 1] (Gnutella website. Http://www.gnutella.com/). In Gnutella, the search is performed by using a method of patroling neighboring nodes using broadcast, so the communication cost at the time of object search becomes a big problem. In order to solve this problem, the paper “A Self-Organizing Access Structure for P2P Information Systems” [Non-Patent Document 2] published by Karl Aberer, Ion Stoica, Robert Morris, Dar. Morris, Dar. A paper published by Franc Kaashhoek and Hari Balakrishnan, “A Scalable Peer-to-Peer Lookup Service for Internet Applications, Non-Patent Document 3”, Sylvia Ratnasulamulak. Content-Addressable Network "[Non-Patent Document 4], Kirsten Hildrum, John D. et al. Kubiatowicz, Satis Rao and Ben Y. et al. Each method is proposed in a paper “Distributed Object Location in a Dynamic Network” published by Zhao [Non-Patent Document 5]. These methods have the following characteristics.

-Efficient object search An efficient search method of O (log (N)) or a search cost according to this is provided for the total number N of nodes. Therefore, the present invention can be applied to an environment where a large number of nodes exist.

-Load equalization Data objects are equally distributed to each node, and the load per node is equalized.

-Support for node addition / deletion Service can be continued by correcting node information autonomously for node addition / deletion.

  However, in these methods, only object search using a key called object ID is considered, and it is not possible to directly perform search using various feature quantities of an object.

Therefore, it is generally considered to use a signature as a means for realizing a flexible search using various feature values of an object. For example, a paper published by Christos Faloutos “Design and Performance Comparison of Signature Extraction Methods” [Non-Patent Literature 6], a paper published by Keiji Ishikawa, Hiroyuki Kitagawa, and Nobuo Ooho's evaluation [Non-Patent Document 7]. By using the signature, it is possible to perform a search using various feature amounts. For example, keyword collation or arbitrary partial character string collation for an object name or its content description can be given.
Gnutella website. http: // www. gnutella. com / “A Self-Organizing Access Structure for P2P Information Systems”, CoopIS 2001, LNCS 2172, pp. 179-194, 2001. “A Scalable Peer-to-Peer Lookup Service for Internet Applications”, SIGCOMM '01, pp. 149-160, 2001. "A Scalable Content-Addressable Network", SIGCOMM '01, pp. 161-172, 2001. “Distributed Object Location in a Dynamic Network”, SPAA '02, pp. 41-52, 2002. “Design and Performance Comparison of Some Signature Extraction Methods” Proc. ACM SIGMOD 1985, pp. 63-82, 1985 “Cost Evaluation of Set Value Retrieval Using Signature File”, Transactions of Information Processing Society of Japan, Vol. 36, no. 2, pp. 383-395, 1995

  However, conventional techniques using signatures have not been studied for use in a P2P environment, and object search using signatures in a P2P environment cannot be performed.

  An object of the present invention is to provide a method for registering and retrieving objects in a P2P environment capable of efficiently searching for objects using various feature quantities, equalizing loads, and adding / deleting nodes, and a program for realizing the method. There is to do.

  Another object of the present invention is to provide an object registration and retrieval method in a P2P environment capable of optimizing the processing efficiency in accordance with the appearance frequency of retrieval requests and object registration requests, and a program for realizing the method. is there.

  Still another object of the present invention is to provide a method for registering and retrieving an object in a P2P environment that can continue object retrieval even when an offline node exists, and a program for realizing this method.

The method of the present invention is a common method using a signature from an object registered in accordance with a common registration method using a signature to a maximum of 2 ⁿ nodes each having a node ID for identifying itself connected to a network. An object registration search method in a P2P environment in which a desired object is searched using a search method is an object.

  In the registration method used in the present invention, at least one feature quantity of an object to be registered is first made an object signature consisting of a bit string of length m (m is a positive integer of 1 or more) (first registration step). Next, the bit string of length m of the object signature is divided into p divided frames composed of a frame signature composed of a bit string of length m / p (p is a positive integer of 1 or more) (second registration) Step). Then, a bit string specifying each of the p divided frames and a bit string of length m / p constituting the frame signature are combined to create a combined signature corresponding to each divided frame (third registration step). Thereafter, p locators composed of a bit string having a length n are generated from the bit string constituting the composite signature (fourth registration step). In the fourth registration step, a locator for specifying a registration destination node is generated using a maximum of n bits from the head of the bit string constituting the composite signature. In this way, the node can be easily and reliably identified by the locator. Next, it consists of the node ID of the node that is performing the registration work, the local ID that identifies the object to be registered within the node that is performing the registration work, the data that identifies the divided frames, and the frame signature. An index entry is created corresponding to p locators (fifth registration step). Finally, this index entry is registered in one or more nodes designated by the locator (sixth registration step).

  In the search method used in the method of the present invention, one or more feature quantities of an object to be searched for at a node performing a search operation are used as a query signature consisting of a bit string of length m (m is a positive integer above) ( First query entry generation step). Next, a bit string of length m of the query signature is divided into p divided frames composed of a query frame signature composed of a bit string of length m / p (p is a positive integer of 1 or more) (second query). Entry generation step). Then, a bit string specifying each of the p divided frames and a bit string of length m / p constituting the query frame signature are synthesized to generate a synthesized signature corresponding to each divided frame (third query entry generation step). . Next, p query locators composed of a bit string of length n are generated from the bit string constituting the composite signature (fourth query entry generating step). The first search step executes these first to fourth inquiry entry generation steps. Next, index frames satisfying the query frame signature condition are collected at one or more nodes determined based on p query locators, and finally query frames collected at one or more nodes from the node that collects the index entries. All index entries that satisfy the signature condition are transmitted to the node that performs the search operation (second search step). When the search method used in the present invention is adopted, index entries that can be used for searching for objects by the first and second search steps can be easily collected in the P2P environment with low communication costs. A method for searching for a target object from the collected index entries may be a manual or a known search method.

  When the index entries created corresponding to the p locators are registered in the nodes designated by the p locators in this way, the index necessary for searching for one object is distributed in the P2P environment and distributed to each node. Will be placed. As a result, the following characteristic advantages can be obtained.

  1. By using a signature, it is possible to cope with a flexible search using various feature amounts.

  2. As the basic framework, Chord, P-Grid, etc., which will be described later, can be used, and the above-described features can be inherited even in a search using a signature.

  3. By introducing the frame division of the signature, by selecting an appropriate parameter, it is possible to adopt a configuration suitable for object search and additional occurrence frequency in the usage environment.

  4). Even when an offline node exists, the object search can be continued.

  First, a search target locator set is generated by obtaining all search target locators that determine nodes to be searched based on the first query locators in the p query locators. All search target locators mean locators for searching for index entries, since index entries may be registered in nodes other than the nodes defined by one query locator. Then, the query data including the node ID of the node performing the search operation and all query frame signatures is sent to the node determined by the first search target locator in the search target locator set (first step). Thereafter, in the first node, all index entries that satisfy the query frame signature condition corresponding to the first search target locator are acquired and set as a partial index entry candidate set (second step). Next, query data and index entry candidates are sent to the next node determined by the next search target locator (third step). At the next node, all index entries satisfying the query frame signature condition corresponding to the next search target locator are acquired, and the union of the index entry set and the partial index entry candidate set is taken and the result is obtained. The next partial index entry candidate set is set (fourth step). Thereafter, the third step and the fourth step are repeated until the last search target locator is reached, and the last partial index entry candidate set is set as an index entry candidate set (fifth step). Also, a search target locator set is generated from the next query locator in the p query locators, and the query data and index entry candidate set are sent to the next node determined by the first search target locator of the search target locator set (No. 6 Step). The index entry candidate set obtained as a result of repeating the second step, the third step, the fourth step, and the fifth step from the next node to the last search target locator of the search target locator set And the product set of index entry candidate sets at the start of this step (seventh step) is taken and the result is set as the next index entry candidate set (seventh step). The sixth step and the seventh step are repeated until the final inquiry locator is reached, and the final result is transmitted to the node that performs the search operation (eighth step).

  By doing so, the search is executed in all the nodes in which the index entry for the predetermined object is registered.

  In each node determined by the query locator, the first query entry generation step to the fourth query entry generation step are executed to determine the next query locator.

The present invention is registered in a computer of a maximum of 2 ⁿ nodes each having a node ID for identifying itself connected to a network according to a common registration procedure using each signature, and a signature is registered from the registered objects. When specified as an object registration search program in the P2P environment used for searching for a desired object using the common search procedure used, this program is configured to execute the following registration procedure and search procedure. . That is, in the registration procedure, a first registration step in which one or more feature quantities of an object to be registered is an object signature composed of a bit string of length m (m is a positive integer of 1 or more), and the length of the object signature. A second registration step of dividing a bit string of length m into p divided frames composed of a frame signature composed of a bit string of length m / p (p is a positive integer of 1 or more), and p divided frames A third registration step for synthesizing a bit string having a length of m / p constituting a frame signature to create a synthesized signature corresponding to each divided frame, and a length from the bit string constituting the synthesized signature A fourth registration step for generating p locators comprising n bit strings, and the node ID of the node performing the registration work First, an index entry corresponding to p locators is created, which includes a local ID for identifying an object to be registered in a node performing registration work, data for identifying a divided frame, and a frame signature. 5 registration steps and a sixth registration step of registering the index entry in one or more nodes designated by the locator.

  The search procedure includes a first query entry generation step in which at least one feature quantity of an object to be searched is set as a query signature consisting of a bit string of length m (m is a positive integer above) in a node performing a search operation. A second query entry that divides a bit string of length m of the query signature into p divided frames composed of a query frame signature composed of a bit string of length m / p (p is a positive integer of 1 or more) Generating a third query entry for generating a composite signature corresponding to each divided frame by synthesizing a generation step, a bit string specifying each of the p divided frames, and a bit string of length m / p constituting the query frame signature Generate p query locators comprising a step and a bit string of length n from the bit string constituting the composite signature Collecting index entries that satisfy the conditions of the query frame signature at one or more nodes determined based on the first query step and p query locators that execute the fourth query entry generation step, and finally index entries A second search step of transmitting all index entries satisfying the conditions of the query frame signature collected at one or more nodes from the node that performs the collection to the node that performs the search operation.

  According to the present invention, it is possible to provide a method for registering and retrieving an object using signature information distributed in a P2P environment. By increasing the number of divided frames, the number of messages at the time of retrieval, the total data transfer amount, Response time can be reduced. Further, according to the present invention, when the number of divided frames is increased, the number of messages when data is added increases, but there is an advantage that there is no significant change in response time due to the introduction of parallel processing. Furthermore, even when the search and additional occurrence rate are taken into account, the processing efficiency can be optimized by changing the number of divided frames. In addition, according to the present invention, with respect to the accuracy of the search when there is an offline node, there is an advantage that a constant search accuracy can be maintained even when there is an offline node, especially when the number of divided frames is increased. It is done.

  Hereinafter, an example of an embodiment of the present invention will be described in detail with reference to the drawings. In the following description, the case where the Chord framework is used as a basic architecture is a specific target. Further, as an example of feature search using a signature, partial character string matching with respect to an object name is shown. Furthermore, through simulation experiments, we evaluate and evaluate the number of messages required when searching and adding objects and their response times, the number of messages when searching and adding are mixed, and the accuracy of searching when there are offline nodes.

  First, before describing the embodiment of the present invention, a structure for searching using a key is introduced into a P2P environment that has been conventionally employed for comparison, and appropriate routing processing is performed for efficient A search method for realizing a simple search will be briefly described. As such a known search method, Chord, P-Grid, CAN, Tapestry, etc. are known. However, in these methods, since the search is limited only to the search based on the object ID as a key, a flexible search based on the feature amount of the object is impossible.

  In the method of the present invention, a signature is used to realize object search in a P2P environment. In addition, the research which makes the collation processing of a signature efficient using parallel processing (Zheng Lin, Concurrent Frame Signature Files, Kluwer Academic Publishers, Distributed and Parallel Databases Vol.1, pp.23-2. In this known method, the signature is divided and equally allocated to all the computers (nodes), and collation processing is performed in parallel to improve efficiency. However, it is very difficult to adapt this known method to a P2P environment because signature verification processing occurs in all nodes and signature division and assignment are static.

  On the other hand, in the present invention, it is possible to reduce the number of nodes that require signature verification processing by devising the signature division method and arrangement method. In addition, dynamic relocation of signatures can be performed even when nodes are added or deleted.

In the embodiment of the present invention, since the Chord architecture is adopted, the Chord architecture will be described first. As shown in FIG. 1, in the Chord architecture, the entire network space is defined as a circular virtual space called an ID circle (Identifier Circle), and all nodes and all data objects are arranged on this ID circle. The size of the Chord network space is represented by a ^scale, and is composed of a maximum of 2 ^scale nodes, that is, 2 ⁿ nodes (n is a positive integer). Chord uses a hash function to evenly arrange nodes and data objects on the ID circle. As a result, a node ID (nid) having a scale bit is given to the node, and an object ID (oid) having a scale bit is given to the data object. Each node is arranged on an ID circle based on the node ID. Further, since a data object having an arbitrary object ID is assigned to any node existing on the ID circle, each node has an ID set.

  The node ID of the node n is n (id), and the node ID of the node n (pre) located immediately before (the clockwise direction is the forward direction) is n (pre) id. At this time, the ID set of the node n is a set of IDs included in the following interval (id).

interval (id) = (n (pre) id, nid)
Each data object is assigned to the node including the object ID as an element of the ID set.

このように、各ｉｎｔｅｒｖａｌの大きさは時計回りに順に２^ｋ−１となっている。これらのルーティング情報を持つことで、各ノードはＩＤサークル全体をカバーすることになり、任意の問合せに対して適切なルーティング処理を行うことができる。 Each node holds a list of actually assigned object IDs, routing information, and node information (successor and predecessor) located before and after. The routing information includes information on a node to which the inquiry should be forwarded (transmitted) next when the object having the object ID to be searched does not exist in the node. This information includes scale, and indicates to which node (referred to as a successor node) the search request should be forwarded next when searching for an object having an object ID included in each interval. However, the interval is a section defined by the following equation.

Thus, the magnitude of each interval is 2 ^k−1 in order clockwise. By having such routing information, each node covers the entire ID circle, and appropriate routing processing can be performed for any query.

  Processing when a node is added or deleted in Chord will be described. When a new node is added, the routing information of the affected node including the node and the node information located before and after the node are updated, and the objects of the successor (next node) are rearranged. When a certain node is deleted, the object of the node is relocated to the successor, and the routing information of the affected node and the node information located before and after are updated.

Next, search processing in Chord will be described in more detail. Chord only considers searches based on object IDs. For this reason, first, an object ID to be acquired is specified in the inquiry. Queries can be started from any node. When a query is passed to a node, the node determines from the list of assigned object IDs whether a data object with that query's object ID exists in the node itself. If it exists, the search is terminated by returning the data object. If the node does not exist, the routing information is checked to determine a node to be forwarded next, and an inquiry is sent to that node. By repeating this series of determination processes, the search based on the object ID is realized. Assuming 2 ^scale nodes, the search space is reduced to half each time a query is forwarded, so the number of messages required to search for a data object is O (log (N)).

  In FIG. 1, the number described in the ellipse is the object ID, and the bit described on the node side indicated by ◯ is the node ID. In addition, the inside of the frame shown on the right side of FIG. 1 includes 2 in the object ID list, and interval (the first ID and the last ID of the object ID to be searched: the former is included, The relationship between the successor (next node) and the successor (next node) and the relationship between the successor (next node) and the predecessor (previous node) are shown. From this relationship, it is understood that the interval is [4, 6], and if the object ID to be obtained is “4”, “5”, the successor (next node) is the node c. If the object ID is '10' to '16', '1', it indicates that the successor (next node) is the node a.

 A specific search example will be described with reference to FIG. As an inquiry, consider searching for an object whose object ID is “2”. First, it is assumed that an inquiry is started from the node c. Since there is no object having the object ID of the query in node c, the query is forwarded to node a which is the successor of [14, 6) including “2” in the interval from the routing information. Similarly, since the object having the object ID of the query does not exist in the node a, the query is forwarded to the node b that is the successor of [2, 4) including “2” in the interval from the routing information. Since there is a data object that matches the query condition at the node b, the result is returned to the start node c, and the search ends. As described above, in Chord, an efficient search is possible for a search using an object ID. Also, since the number of elements in the ID set of each node is almost the same, the load can be equalized. Furthermore, correspondence to addition and deletion of nodes is also made. Therefore, the above-mentioned three features are realized.

  Next, the signature used in the present invention will be described. The signature is a fixed-length bit string generated from each data object and expresses the feature quantity of the object. By converting the feature quantity of an object into a simple expression method called a bit string, the presence or absence of a specific feature quantity can be easily determined, and object search using various feature quantities becomes possible.

  First, an object signature generated from each data object will be described. FIG. 3 shows an example in which a known trigram generated from the name of a data object such as a file name is used as a feature amount (characters constituting the name are divided into three characters and used as feature amounts). . Well-known superimposed coding is used to generate an object signature including a bit string. In this method, each feature quantity is hashed to generate a signature-length element signature (pur, ure, reP, eP2, and P2P represented by a bit string). Further, a logical sum of each bit string of the obtained element signature is taken to make the bit string an object signature. The object signature generated in this way represents all the feature quantities of the data object. In the signature, this is expressed by the bit position where “1” stands. The number of bits in which “1” stands is called a signature weight.

上記式においては、Ｎｕｍ＿ＦａｌｓｅＤｒｏｐは、フォールスドロップ数、Ｎｕｍ＿ＵｎｑｕａｌｉｆｉｅｄＤａｔａＯｂｊｅｃｔは、問合せ条件を満たさないデータオブジェクト数である。このように、ｔｒｉｇｒａｍを特徴量と考えた場合には、シグネチャを用いることで、任意の文字列によるデータオブジェクト名の部分一致検索を実現できる。 As for the query, as shown in FIG. 4, a query signature is created in the same manner as the object signature of FIG. The example of FIG. 4 shows an example of creating a query signature with the name of the object to be searched as “pure”. In this example, pure is decomposed into a trigram, and a query signature is created by taking the logical sum of the element signatures of each feature quantity. The bit position where “1” stands in the query signature represents all the feature values of the query. In the object search, by checking whether the bit position where “1” is set in the query signature is also “1” in the object signature, it is determined whether there is a possibility that the object includes all the features of the query. judge. Therefore, when the logical product of each bit of the object signature and the query signature matches the query signature, the object may contain all the features of the query, and compensation of the solution that satisfies the query condition. It becomes. Compensation of this solution is called drop, the actual answer in this is called actual drop, and the other is called false drop. This determination process is called false drop resolution. Further, the probability that the solution compensation is a false drop is called a false drop probability Fd, which is given by the following equation. False drop probability is a measure for measuring signature search accuracy.

In the above formula, Num_FalseDrop is the number of false drops, and Num_UnqualifiedDataObject is the number of data objects that do not satisfy the query condition. As described above, when the trigram is considered as the feature amount, the partial match search of the data object name using an arbitrary character string can be realized by using the signature.

  In the present invention, the object information based on various feature quantities is realized by distributing signature information on nodes of the P2P network. As the simplest distributed arrangement method, it can be considered to distribute and arrange each object signature as one unit. Although this method can efficiently process an object when it is added, it may be necessary to refer to a large number of signatures registered in each node at the time of searching. difficult. As another distributed arrangement method, each object signature may be divided into a plurality of parts and distributedly arranged in units of partial signatures. In this case, since only the partial signature including the bit position where “1” stands in the query signature needs to be referred to at the time of collation, it is possible to improve the efficiency of the search process. However, when adding an object, it is necessary to place each partial signature. Therefore, in the present invention, the above-described distributed arrangement method using the object signature as a unit and the distributed arrangement method using the partial signature as a unit are combined to realize a method capable of constructing the object according to the usage environment and the additional occurrence frequency. did. The method of the present invention is built on a framework that realizes efficient data retrieval on a P2P network, and can be implemented using any of the above-described frameworks such as Chord and P-Grid. Is possible. In the present embodiment, Chord is used.

FIG. 5 is a flowchart showing an algorithm of registration processing in a program used for carrying out the embodiment of the method of the present invention, and FIG. 6 is a flowchart showing an algorithm of search processing in this program. Hereinafter, an example of an embodiment of the method and the program of the present invention will be described while explaining the steps of FIGS. The programs that implement the algorithms shown in FIGS. 5 and 6 are all installed on computers in a maximum of 2 ⁿ nodes (n is a positive integer) having node IDs that identify themselves connected to the network in the P2P environment. ing. The function shown in FIG. 5 means the following. len () means returning the length of the array or bit string given by the argument. value () means that a value obtained by converting an argument into a decimal representation is returned. binary () means returning a value obtained by converting the argument into a binary representation. “concatenate ()” means that a bit string obtained by concatenating the first argument and the second argument is returned. prefix () means that a bit string from the head to the second argument is returned for the bit string of the first argument. Zero padding () means that the bit string of the first argument is padded with “0” until the length of the bit string becomes the second argument, and the generated bit string is returned. In FIG. 6, check () means that an index entry set that satisfies the following two conditions 1 and 2 is returned for the index entry set given by the first argument.

1. 1. Frame number of each entry = value of first argument Frame signature of each entry Λ Signature of third argument = Signature of third argument (Λ represents bitwise AND)
Next () means that a frame signature next sig satisfying the following two conditions 1 and 2 is calculated based on the frame signature portion of the locator given by the argument, and a new locator is generated and returned. .

1. next sig Λ frame signature part = frame signature part next (next sig)> next sig that is the smallest value (next sig) that satisfies value (frame signature part)
First, in the registration process, in step ST1, as shown in FIG. 5, the user places the data object to be searched at an arbitrary node in order to execute the data object addition process in the node performing the registration work. Then, according to steps ST2 to ST15, index entries including signature information are distributed and arranged at predetermined nodes. Index entry placement processing and search processing use the Chord framework. Since each data object is managed in node units, a pair of an ID in the node of the data object and a node ID storing the data object is a key for uniquely determining the data object. In order to clearly distinguish the Chord object ID (oid) from the ID of the data object of the present invention, the ID of the data object in the node in this embodiment will be referred to as a local ID (lid).

  In steps ST2 and ST3, preprocessing for performing distributed arrangement of signature information is performed. First, in step ST2, an object signature is generated by the method described with reference to FIG. In step ST3, the object signature generated from the data object is divided into the number of divided frame = slice frame signatures as shown in FIG. 7 (second registration step) (“Frame-Sliced by Zheng Lin and Christos Faloutos”). Signage Files "IEEE TKDE Vol. 4, No. 3, pp. 281-289, 1992.). The divided frame number P is a positive integer of 1 or more. In the example of FIG. 7, when the number of divided frames is 4, and a 16-bit data signature or object signature (query signature in the case of search), four divided frames (f0 to f3) are generated, and each divided frame is generated. Are each constituted by a frame signature composed of a 4-bit bit string. When the number of divided frames is 1, the frame signature matches the object signature. In particular, the case where slice matches the signature length is called a bit slice configuration.

  Next, in step ST4, each frame signature is numbered. Then, a combined signature is obtained by combining the bit string obtained by converting the divided frame number into binary representation in the following steps ST5 to ST12 and the frame signature of the divided frame (third registration step). In the present embodiment, p locators having a length of scale bits (n = 4 bits) are generated from the composite signature. The p locators are used for the arrangement of index entries on the ID circle and the search process. The locator generation method is as follows. First, as shown in the upper right part of FIG. 7, a combined signature corresponding to each divided frame is generated by combining a bit string that identifies each divided frame with i + fi and a bit string of length m / p that constitutes the frame signature. (Step ST7). Then, a locator is generated by the highest n bits of the composite signature (steps ST8 to ST12). In step ST5, it is determined whether or not the processing has been completed for all of the divided frames. If all the processing has been completed, the processing ends. If all the processes have not been completed for the divided frames, the process proceeds to step ST6. If the bit strings constituting the frame signature are all composed of 0, the process proceeds to step ST15 and the next divided frame is processed. If at least one 1 is included in the bit string constituting the frame signature, the process proceeds to step ST7 to generate a composite signature. Depending on the length of the bit string of the generated composite signature, the processing from step ST8 to step ST12 is executed to generate a locator. The following processing is performed in steps ST8 to ST11.

  [Case 1] When the combined signature length is the same as the scale (case 4 in FIG. 7), the combined signature itself is used as a locator (steps ST8, 10, 12).

  [Case 2] When the composite signature length is larger than scale (larger than 4 in FIG. 7), the scale (up to the fourth in FIG. 7) prefix (bits from the beginning of the bit string) from the head is used as the locator. (Steps ST8, 9).

  [Case 3] When the combined signature length is smaller than the scale (less than 4 in the case of FIG. 7), the length is scaled by adding “0” to the combined signature (the number of bits in FIG. 7 is 4). A locator is used (steps ST8, 10, 11).

ロケータの順序は上記式のｏｒｄの昇順とする。 By using this locator generation method, a locator can be generated for an arbitrary number of divided frames and signature length (fourth registration step). Next, proceeding to step ST13, an index entry is generated. In step ST13, the node ID of the node that is performing the registration work, the local ID that identifies the object to be registered in the node that is performing the registration work, the data that identifies the divided frames, and the frame signature Are created in correspondence with the corresponding locators (fifth registration step). In step ST14, the index entry is arranged or registered in one or more nodes designated by the locator (sixth registration step). By arranging the index entries using such a locator, the index entries can be arranged equally for each node. Here, the order of the locators in the node ni is defined as follows. When the node ID of the node ni is nid and the locator is loc, the next ord is calculated.

The order of the locators is the ascending order of ord in the above formula.

  Next, the generation of index entries and the arrangement method thereof will be described in more detail. An index entry is generated for each frame signature. However, no index entry is generated for a frame signature in which all bits are “0”. The index entry includes a local ID of the data object, a node ID storing the data object, a frame number, and a frame signature thereof.

  An arrangement method of index entries when a data object in a certain node is added will be described. In this embodiment, the index entry placement processing is performed in parallel to shorten the placement processing response time. First, a locator is generated from the frame number and frame signature in each index entry. Then, as shown in FIG. 8, each index entry is arranged at an appropriate node according to the Chord algorithm, regarding the locator as an object ID. At this time, instead of individually processing each index entry that needs to be arranged, first, each index entry is set for each node to be forwarded based on the routing information of the node to which the data object is added. Classify. Next, each index entry set obtained in this way is sent to the node. Similar processing is repeatedly performed at the node to which the index entry is sent, and index entries are arranged. In this arrangement method, the number of messages for index entry arrangement processed by each node can be minimized. In the following, an index entry set arranged by a certain locator Loc is referred to as a Loc arrangement index entry set.

  Next, the index entry search method for searching for an object will be described with reference to FIG. Also in the object search, a query signature, a frame signature, a composite signature, and a locator are sequentially generated from the feature amount given as the query condition. However, no locator is generated for a frame signature composed entirely of “0”. These are realized by step ST21 to step ST31 in FIG. 6 (first search step). Steps ST21 to ST31 are substantially the same as steps ST1 to ST13 in FIG. Specifically, at the node that performs the search operation, one or more feature quantities of the object to be searched are generated as a query signature consisting of a bit string of length m (m is a positive integer above) (first query entry) Generation step ST21). Then, the bit string of length m of the query signature is divided into p divided frames composed of a query frame signature composed of a bit string of length m / p (p is a positive integer of 1 or more) (second query). Entry generation step ST22). Next, a bit string specifying each of the p number of divided frames and a bit string of length m / p constituting the query frame signature are synthesized to generate a synthesized signature corresponding to each divided frame (third query entry generation) Step ST26). Then, p query locators composed of n bit strings are generated from the bit strings constituting the composite signature (fourth query entry generating steps ST27 to ST31). This first search step is initially executed at the node that performs the search operation. Thereafter, step ST24 to step ST31 (the first query entry generation step to the fourth query entry generation step described above) are executed at each node determined by the previous query locator to determine the next query locator.

  In step ST32, a set of intermediate results is set to an initial value at the first node. That is, the storage location is made empty. In subsequent steps ST33 to ST41, acquisition and narrowing down of the index entry set is performed as follows. First, in step ST33, a search target locator set is generated by obtaining all search target locators that determine nodes to be searched based on the first query locators in the p query locators (steps ST36 and 37). Query data including the node ID of the node that performs the work and all query frame signatures (ie, partial query entries) is sent to the node determined by the first search target locator in the search target locator set (step ST33: first step). Then, the process proceeds to step ST34, and at the first node, all index entries that satisfy the query frame signature condition corresponding to the first search target locator are acquired and set as a partial index entry candidate set (second step). . Next, inquiry data and index entry candidates are sent to the next node determined by the next search target locator (steps ST36 and 37: third step). Then, in the next node, all index entries that satisfy the query frame signature condition corresponding to the next search target locator are acquired (step ST33), and the union of the index entry set and the partial index entry candidate set is taken. The result is set as the next partial index entry candidate set (steps ST34 and 35: fourth step). The third step and the fourth step are repeated until the last search target locator is reached, and the last partial index entry candidate set is set as an index entry candidate set (fifth step). These first to fifth steps are executed in steps ST33 to ST37 in FIG. In step ST36, since the locators are calculated sequentially, the end condition is when all the frame signature portions of the locators are "1".

  In step ST38, it is determined whether or not the process is the first query locator process. If the process is the first query locator process, the process proceeds to step ST39 and the collected index entry candidates are directly used as the next query locator. (Steps ST40, ST42, ST24 to ST31).

  Next, a search target locator set is generated from the next query locator in the p query locators, and the query data and index entry candidate set are sent to the next node determined by the first search target locator of the search target locator set (step ST39 and ST40: 6th step). The above-mentioned second step, third step, fourth step and fifth step are repeated until the last search target locator of the search target locator set is reached from the next node, and index entry candidates obtained as a result thereof The product set of the set and the index entry candidate set at the start of this step (step ST41) is taken and the result is set as the next index entry candidate set (seventh step: step ST24 to step ST38, step ST41). Then, the sixth step and the seventh step are repeated until the final inquiry locator is reached, and the final result is transmitted as a solution to the node performing the search operation (eighth step: steps ST24 and ST43).

  The partial query entry described above includes a locator, a frame number, and a frame signature. Queries are processed by sequentially matching these partial query entries. This processing is performed by sequentially circulating the nodes having index entries to be referred to when performing the query processing in the clockwise direction. In this case, since narrowing down of solutions by the collation processing is sequentially performed, it is expected that the data transfer amount of the intermediate result is reduced.

FIG. 9 shows a search algorithm that separately expresses the first and second search steps. First, a partial query entry set is generated from the query (first line). Each partial query entry is processed in order of having a smaller locator according to the locator order in the node where the query occurs (second line). Next, a search target locator set is generated from the locators of the partial query entries (third line). The search target locator set is a set of all bit strings in which “0” in the frame signature in the locator is set to “0” or “1” (thus satisfying the following expression. Each element Λ of the search target locator set) ^* (Λ represents bitwise AND) Locator = Locator) At this point, no search processing is performed on the partial query entry, so the solution complement set is an empty set (line 4). However, the solution complement set is a set having a pair of a node ID and a local ID as an element. Next, for each element of the search target locator set, the node that holds the placement index entry set of the locator is set to the current solution complement set, the unprocessed partial query entry set, and the unprocessed search target locator set. Will be sent. However, the locator selection from the search target locator set follows the locator order in the node. The node that receives them selects an index entry satisfying the following two conditions from the set of arrangement index entries held by itself (line 6), and resolves the pair of the node ID and local ID in the index entry. Add to complement (7th line).

  [Condition 1] Frame number in partial query entry = frame number in index entry.

  [Condition 2] Frame Signature in Partial Query Entry Frame Signature in Λ Index Entry = Frame Signature in Partial Query Entry

  When the acquisition process of the solution complement set is completed for each element of all search target locator sets, the process related to the partial query entry ends. Thereafter, the product of the set of the solution Ans and the solution complement set calculated so far is taken, and the solution narrowing process is performed (line 11). After performing this processing for all partial query entries, Ans is returned to the start node (line 12).

The inquiry in FIG. 10 will be described as an example. First, it is assumed that an inquiry occurs at the node n4. Query processing in the order of the locator of the node n4, it is started from the partial query entry d _0. A search target locator set d ₀ Set {'0010', '0011'} is generated from d _0, and the presence / absence of an arrangement index entry set of '0010' is checked. Node n4 has no arrangement index entry set of “0010”. For this reason, by using the routing information, the node n1 where the arrangement index entry set of “0010” exists is traced. At this time, the unprocessed partial query entry set {d ₀ , d ₂ }, the unprocessed search target locator set d ₀ Set, and the solution complement set are sent to the node n1. Since the node n1 has an arrangement index entry set of all locators ('0010' and '0011') in d ₀ Set, a solution complement set using these is obtained. At this point, the processing for all locators in d ₀ Set is complete, so the processing for d ₀ is complete. Then untreated Rates partial query entry _{d 2} of the object locator set _{d 2 Set { '1000',} '1010', '1001', '1011'} is generated, the presence or absence of the arrangement index entry set of '1000' Investigate. Node n1 does not have a placement index entry set of “1000”, and uses routing information to trace node n3 where the placement index entry set exists. At this time, the unprocessed partial query entry set {d ₂ }, the unprocessed search target locator set d ₂ Set, and the solution set Ans are sent to the node n3. In node n3, there is an arrangement index entry set of all locators in d ₂ Set, and the processing for d ₂ ends. Here, Ans is recalculated. At this stage, the signature verification process ends, and Ans as a search result is returned to the start node n4.

  The start node n4 obtains data objects corresponding to all elements in the search result Ans, and finally performs false drop resolution to obtain a final solution for the query.

シグネチャ長およびウェイトについては、表２に示す３通りについて実験を行う。データオブジェクトの特徴量の数を１６４とした場合、特徴量１個に基づく問合せに対するフォールスドロップ確率はこの３通りのいずれの場合も約５％となる。応答時間の測定では、ノード間で、あるメッセージを転送するために必要な転送時間を単位時間とする。その他の処理時間は、メッセージ転送に必要な時間と比較した場合に非常に小さくなると考えられるため、考慮しない。 Hereinafter, an experiment evaluating the above embodiment will be described. In the experiment, an evaluation experiment of the method of the present embodiment based on simulation was performed. In the experiment, 1) number of messages required for search, total data transfer amount, response time, 2) number of messages and response time for index entry update accompanying object addition, and 3) search and addition are mixed. 4) Measure the accuracy of the search when there is an offline node. As described above, there is a trade-off relationship between the search cost and the additional cost, and the number of divided frames is a parameter that greatly affects the cost. Confirming this relationship is the main purpose of Experiments 1), 2) and 3). Further, as one of the features of search using signatures, there is a point that it is possible to narrow down target objects without referring to all frame signatures. With this feature, even if a node that holds a necessary index entry is offline, it is possible to perform a narrowing process using only the index entries of the remaining online nodes. The purpose of Experiment 4) is to confirm this point. Tables 1 and 2 show the main parameters during the experiment.

For the signature length and weight, the experiment is performed for three types shown in Table 2. When the number of feature amounts of the data object is 164, the false drop probability for a query based on one feature amount is about 5% in any of these three cases. In the measurement of response time, the transfer time required for transferring a certain message between nodes is defined as a unit time. Other processing times are not considered because they are likely to be very small when compared to the time required for message transfer.

[Search Object]
First, the number of messages, total data transfer amount, and response time when searching are evaluated. The experiment is performed by changing the number of divided frames slice and the signature length F. The number of nodes is 128, and the feature amount of the query is changed to 2, 4, and 6. As already described, it is expected that the search cost decreases as the number of divided frames increases. Further, when the signature length is increased, the weight of the query signature is reduced, so that the number of frame signatures that need to be referred to is reduced and the search cost is expected to be reduced. On the other hand, when the number of query feature values is increased, the weight of the query signature also increases, and the search cost is considered to increase. In the experiment, an average when the query is executed 50 times is used as a measurement value. When measuring the data transfer amount, the size of each element of the index entry is set to nid is 6 [Byte] and lid is 4 [Byte]. The size of the frame number and the size of the frame signature vary depending on the value of slice. When the number of bits necessary to express the slice value is bit_slice, the size required for the frame number is [bit_slice / 8] [Byte] ([] represents the smallest integer greater than or equal to x), The required size is [F / bit * slice] [Byte].

  FIG. 11 shows the average number of messages. The number of messages decreases as the number of divided frames increases. This results in a reduction in the number of messages because when a query signature is divided, a frame signature (a frame signature composed of all “0”) that does not need to determine a solution appears with a high probability. Furthermore, since the frame signature length is also reduced, the number of elements in the search target locator set is reduced, and the number of nodes that must be traced is also reduced. For these reasons, the number of messages is greatly reduced.

There is a similar tendency with respect to the total data transfer amount (FIG. 12). The case without frame division (data transfer amount to the divided frame number 2 ⁰ is small, the entire object signature of the data object only in a single index entry is obtained, due to the fact that narrowing of the solution can be performed instantaneously. Division the number of frames is increased the total amount of data transfer in the vicinity of 2 ¹ or more, increased data transfer amount of the intermediate results for can not be performed narrowing of sufficient de by sequential cyclic, even if the number of messages required to search This is because it is getting bigger.

  The response time will be described. In the proposed method, the search is performed by sequentially visiting each node, and the result is returned when all the narrowing-down processes are completed. Therefore, the response time has the same curve as the number of messages at the time of search (FIG. 11). Also in this case, the case where the number of divided frames is increased is superior.

  From the experimental results, it can be seen that the larger the number of divided frames, the more efficient the number of messages at the time of retrieval. It can also be confirmed that the number of messages can be reduced by reducing the signature weight.

[Add Object]
Next, when a data object is newly added, the number of messages necessary for the arrangement of index entries and the response time are measured. In the experiment, the number of divided frames and the signature length F are changed, and the number of nodes is changed to 128 and 256. The arrangement method, that is, the registration method is performed based on the method described above. As described above, it is expected that the number of messages required for addition increases as the number of divided frames increases. Similarly to the object search, it is expected that the number of messages required for the arrangement of the index entry is reduced when the signature length is increased.

  FIG. 13 shows the average number of messages. It can be seen that the larger the number of divided frames, the larger the number of messages. This is because as the number of divided frames increases, the number of index entries that need to be newly arranged increases. When the signature weight is reduced, the number of messages is relatively small. This is because the number of index entries that need to be arranged is reduced.

  The response time is also measured. FIG. 14 shows the experimental results. Since the index entries are arranged in parallel, the response time increase curve is gradual with respect to the increase in the number of index entries due to the increase in the number of divided frames. It can also be seen that the response time hardly changes for any signature length.

  As described above, when the number of divided frames is increased, the number of messages tends to be very large. However, when viewed in response time, there is not much difference compared to the case where frame division is not performed.

[When search and addition are mixed]
Consider the average number of messages considering the frequency of object search processing and object addition processing. Here, it is assumed that the occurrence probability of the search process is p, and the occurrence probability of the additional process is (1-p). In the experiment, the number of nodes is 128, the divided signature and the signature length are changed, and the average number of messages is obtained by multiplying the number of messages at the time of retrieval and the number of messages at the time of addition by the respective occurrence probabilities.

FIG. 15 shows the experimental results. In general, it is considered that search processing occurs more frequently than addition processing, so p is changed to 0.7, 0.8, and 0.9. For p = 0.7, the number of divided frame message number becomes the smallest at around 2 ^4. In addition, when p = 0.8, the number of messages is the smallest when the number of divided frames is around ²⁵ , and when p = 0.9, the number of divided frames is ²⁵ or more and the minimum It can be seen that the number of messages is almost constant. Thus, the optimum number of divided frames varies depending on the search and the additional occurrence rate.

[Accuracy of search when offline node exists]
Finally, an experiment is performed on the accuracy of search when it is assumed that a predetermined percentage of nodes are offline and signature verification processing cannot be performed correctly. In the Chord framework, routing information can be dynamically updated even when an offline node exists. For this reason, it is assumed that the routing information can always be used correctly. In this experiment, the false drop probability when a query is executed is calculated. When offline nodes exist, the frame signatures existing in these offline nodes cannot be collated, so the number of false drops is expected to increase.

In the experiment, the number of query feature amounts is set to 1, and the number of divided frames slice is changed to 2 ⁰ , 2 ⁴ , and 2 ⁸ to determine the false drop probability. Further, the number of nodes is 128, and the ratio of nodes that cannot be correctly verified is changed by 2.5% from 0% to a maximum of 50%.

  FIG. 16 shows the experimental results. It can be confirmed that the false drop probability increases as the proportion of offline nodes increases. When the number of divided frames is increased, the increase curve of the false drop probability is gentler. The reason for this is considered that the size of the frame signature is reduced by increasing the number of divided frames, and the size of the signature that cannot be verified is reduced. In this way, even in the situation where offline nodes exist, a certain level of search accuracy can be maintained by increasing the number of divided frames.

It is a figure used in order to explain Chord architecture. It is a figure used in order to explain the search processing by Chord. It is a figure used in order to explain the generation method of an object signature. It is used to explain how to generate a query signature. It is a flowchart which shows the algorithm of the registration process in the program used in order to implement embodiment of the method of this invention. It is a flowchart which shows the algorithm of the search process in a program. It is a figure for demonstrating the production | generation method of a locator. It is a figure used in order to demonstrate arrangement | positioning of an index entry. It is the figure which tabulated the search algorithm. It is a figure used in order to explain object search. It is a figure which shows the experimental result about the average number of messages at the time of a search. It is a figure which shows the experimental result about the average total data transfer amount at the time of a search. It is a figure which shows the experimental result about the average number of messages at the time of arrangement | positioning. It is a figure which shows the experimental result about the average response time at the time of arrangement | positioning. It is a figure which shows the experimental result about the number of messages in case search and addition coexist. It is a figure which shows the experimental result about the false drop probability when there exists an offline node.

Claims (7)

Common using signatures from objects registered according to a common registration method using signatures to a maximum of 2 ⁿ nodes (n is a positive integer) connected to the network and having node IDs identifying themselves. An object registration search method in a P2P environment for searching for a desired object using the search method of
The registration method is:
A first registration step in which one or more feature quantities of an object to be registered is an object signature consisting of a bit string of length m (m is a positive integer of 1 or more);
A second registration step of dividing the bit string of length m of the object signature into p divided frames composed of a frame signature composed of a bit string of length m / p (p is a positive integer of 1 or more); ,
A third registration step of synthesizing a bit string specifying each of the p number of divided frames and a bit string of length m / p constituting the frame signature to create a combined signature corresponding to each divided frame;
A fourth registration step of generating p locators comprising a bit string of length n from the bit string constituting the composite signature;
A node ID of the node that is performing the registration work, a local ID that identifies an object to be registered in the node that is performing the registration work, data that identifies the divided frame, and the frame signature A fifth registration step of creating an index entry corresponding to the p locators;
A sixth registration step of registering the index entry with one or more of the nodes specified by the locator;
The search method is:
A first query entry generating step for generating one or more feature quantities of an object to be searched as a query signature consisting of a bit string of length m (m is a positive integer of 1 or more) in the node performing the search operation; A second query entry that divides the bit string of length m of the query signature into p divided frames composed of a query frame signature composed of a bit string of length m / p (p is a positive integer of 1 or more). A third query for generating a combined signature corresponding to each divided frame by combining a generating step, a bit string specifying each of the p divided frames, and a bit string of length m / p constituting the query frame signature An entry generation step and p queries comprising a bit string of length n from the bit string constituting the composite signature A first search step of performing a fourth query entry generation step of generating an applicator,
Index entries that satisfy the query frame signature condition are collected at one or more nodes determined based on the p query locators, and finally collected at the one or more nodes from the node that collects index entries. A method for registering and retrieving an object in a P2P environment, comprising: a second retrieval step of transmitting all the index entries that satisfy the condition of the inquiry frame signature to a node that performs the retrieval operation.   2. The object registration search according to claim 1, wherein, in the fourth registration step, the locator for specifying a registration destination node is generated using a maximum of n bits from the top of a bit string constituting the composite signature. Method. The second search step includes
Based on the first query locator among the p query locators, all search target locators that determine the node to be searched are obtained to generate a search target locator set, and a node ID of a node that performs the search operation and A first step of sending query data including all the query frame signatures to the node defined by the first search target locator of the search target locator set;
A second step of obtaining all index entries satisfying the condition of the query frame signature corresponding to the first search target locator as the partial index entry candidate set in the first node;
A third step of sending the query data and the index entry candidate to the next node determined by a next search target locator;
In the next node, all index entries that satisfy the conditions of the query frame signature corresponding to the next search target locator are acquired, and the union of the index entry set and the partial index entry candidate set is obtained. A fourth step in which the result is the next partial index entry candidate set;
A fifth step of repeating the third step and the fourth step until the last search target locator is reached, and setting the last partial index entry candidate set as an index entry candidate set;
The search target locator set is generated from the next query locator in the p query locators, and the query data and the index entry candidate set are determined by the first search target locator of the search target locator set. A sixth step to send to
The second step, the third step, the fourth step, and the fifth step are repeated as a result, starting from the next node until reaching the last search target locator of the search target locator set. A seventh step of taking the product set of the index entry candidate set and the index entry candidate set at the start of this step and setting the result as the next index entry candidate set;
In the P2P environment, the sixth step and the seventh step are repeated until the last inquiry locator is reached, and the final result is transmitted to the node that performs the search operation. How to register and search for objects.   3. The P2P according to claim 2, wherein the first query entry generation step to the fourth query entry generation step are executed to determine the next query locator at each of the nodes determined by the query locator. How to register and search for objects in the environment. Register to a computer of up to 2 ⁿ nodes each having a node ID that identifies itself by being connected to the network according to a common registration procedure using a signature, and using a signature from a registered object. A program for registering and retrieving objects in a P2P environment used for retrieving a desired object using a retrieval procedure,
The registration procedure includes:
A first registration step in which one or more feature quantities of an object to be registered is an object signature consisting of a bit string of length m (m is a positive integer of 1 or more);
A second registration step of dividing the bit string of length m of the object signature into p divided frames composed of a frame signature composed of a bit string of length m / p (p is a positive integer of 1 or more); ,
A third registration step of synthesizing a bit string specifying each of the p number of divided frames and a bit string of length m / p constituting the frame signature to create a combined signature corresponding to each divided frame;
A fourth registration step of generating p locators comprising a bit string of length n from the bit string constituting the composite signature;
A node ID of the node that is performing the registration work, a local ID that identifies an object to be registered in the node that is performing the registration work, data that identifies the divided frame, and the frame signature A fifth registration step of creating an index entry corresponding to the p locators;
A sixth registration step of registering the index entry with one or more of the nodes specified by the locator;
The search procedure is
A first query entry generating step in which at least one feature quantity of an object to be searched is a query signature consisting of a bit string of length m (m is a positive integer of 1 or more) in the node performing the search operation; Generation of a second query entry that divides the bit string of length m of the query signature into p divided frames composed of a query frame signature consisting of a bit string of length m / p (p is a positive integer of 1 or more) A third query entry for synthesizing a bit string having a length of m / p and a bit string specifying each of the p divided frames and a bit string of length m / p constituting the query frame signature And p query locators comprising a bit string of length n from the bit string constituting the composite signature A first search step of performing a fourth query entry generation step of generating,
Index entries that satisfy the query frame signature condition are collected at one or more nodes determined based on the p query locators, and finally collected at the one or more nodes from the node that collects index entries. A second search step of transmitting all the index entries that satisfy the condition of the query frame signature to a node that performs the search operation,
An object registration / retrieval program in a P2P environment for executing the registration procedure and the retrieval procedure. The second search step includes
Based on the first query locator among the p query locators, all search target locators that determine the node to be searched are obtained to generate a search target locator set, and a node ID of a node that performs the search operation and A first step of sending query data including all the query frame signatures to the node defined by the first search target locator of the search target locator set;
A second step of obtaining all index entries satisfying the condition of the query frame signature corresponding to the first search target locator as the partial index entry candidate set in the first node;
A third step of sending the query data and the index entry candidate to the next node determined by a next search target locator;
In the next node, all index entries that satisfy the conditions of the query frame signature corresponding to the next search target locator are acquired, and the union of the index entry set and the partial index entry candidate set is obtained. A fourth step in which the result is the next partial index entry candidate set;
A fifth step of repeating the third step and the fourth step until the last search target locator is reached, and setting the last partial index entry candidate set as an index entry candidate set;
The search target locator set is generated from the next query locator in the p query locators, and the query data and the index entry candidate set are determined by the first search target locator of the search target locator set. A sixth step to send to
The second step, the third step, the fourth step, and the fifth step are repeated as a result, starting from the next node until reaching the last search target locator of the search target locator set. A seventh step of taking the product set of the index entry candidate set and the index entry candidate set at the start of this step and setting the result as the next index entry candidate set;
In the P2P environment, the sixth step and the seventh step are repeated until the last inquiry locator is reached, and the final result is transmitted to the node that performs the search operation. Object registration search program.   The first query entry generation step to the fourth query entry generation step are executed at each of the nodes determined by the query locator to further realize a function of determining the next query locator. 7. A program method for registering and retrieving objects in the P2P environment according to 6.

Images